Modeling highly migratory species in data-deficient frontier zones: a targeted pseudo-absence selection framework

In rapidly changing marine environments, reactive management strategies based on long-term ecological monitoring are often insufficient for timely conservation actions. This limitation is particularly pronounced in transitional or data-deficient frontier zones, where ecological conditions shift rapidly and systematic field surveys remain sparse. Modeling endangered and highly mobile marine megafauna is especially challenging due to the scarcity of occurrence data and uncertainty regarding absence information. Using the whale shark (Rhincodon typus) in Korean waters (n = 33) as a case study, we developed an integrated species distribution modeling (SDM) framework that combines spatial expansion of the modeling domain with a Long-term Inverse-Weighted (LIW) pseudo-absence generation method. The LIW approach assigns greater sampling weights to areas that have been unrecorded for extended periods, thereby enhancing the spatial and environmental distinction between presence and absence data. To evaluate its performance, we compared LIW with uniform random and seasonal inverse-weighted methods across four algorithms: boosted regression trees, random forest, extreme gradient boosting, and maximum entropy. While classification metrics improved only modestly, the LIW-driven refinement yielded spatially distinct and ecologically coherent habitat predictions, especially from the tree-based models. These LIW-based models successfully identified the northern edge of the whale shark’s seasonal range along the southern and eastern coasts of Korea. This study shows that methodological integration, combining targeted pseudo-absence selection with spatial expansion using open-access data, can support more reliable SDM in data-limited regions, offering a practical framework for proactive marine conservation under data scarcity.